Vortrag: Parallelism across Architectures

https://e.feu.de/zoommi
Meeting-ID: 339 339 4488

In this talk, I will explore modern challenges and opportunities in parallel programming with a focus on two architectures: GPUs and RISC-V CPUs. On the GPU side, I investigate Python-based programming approaches for (multi)-GPU systems, discussing both their practical potential and the limitations imposed by Python’s dynamic nature — particularly in the context of device communication, kernel management, and runtime overhead. I will also offer a perspective on
teaching this topic, addressing the challenges and limitations of using Python for GPU computing — while highlighting its potential to lower entry barriers and motivate more students to engage with parallel programming. On the RISC-V side, the talk highlights the vector extension as a promising building block for parallelism in embedded and scientific applications. I will examine its integration into MPI and OpenMP workflows and explore different application domains where RISC-V vector
capabilities may offer performance.